We have some new hardware and are running into a problem with the symbol error counter when we run 1 port in IB mode and 1 in ethernet. The symbol error counter for the IB side pegs to 65532 and does not reset as soon as the port changes mode to ethernet. I have tried 2 different drivers 4.3-1 and 4.3-3 both of which exhibit the same behavior. The device is DEL1090001019 (firmware 2.42.5000) running in a Dell C4140. OS is OL 7.5. We are also seeing this same behavior in an R730 with DEL1090001019 (firmware 2.36.5000) pm CentOS 7.4. Has anyone else run into this and know of a workaround or a way to fix it? The nodes always appear when running ibqueryerrors on the fabric when trying to find bad cables. We do have hardware running the same configuration with ethernet and IB in this configuration with no problems, it seems like it is just Dell hardware experiencing this problem.
Usually, symbol errors caused by some physical condition and in many cases fixed by a) reseating BOTH ends of the cable or b) replacing the cable. If you are using OEM solution, you might contact the hardware vendor after trying reseating the cables and see if your equipment is under the warranty or open case with him.
In order to reset the fabric counters, use ‘ibdiagnet -pc’ command and the same command should be used to collect information about the fabric. ibqueryerrors, despite that it exists in Mellanox OFED, shouldn’t be used as it not under the development. ibdiagnet is a swiss army knife.